On Cohesion and How It Actually Shows Up in Real Texts
Cohesion is the network of linguistic links that hold a text together. That sounds like something you'd read in a textbook. The practical version is simpler: it's why a paragraph doesn't fall apart when you read it. You know immediately that "it" refers to the noun from two sentences back, and you don't have to stop and reconstruct the logic. Halliday and Hasan, in their 1976 work, mapped out the mechanics of this. They identified the main types and showed how they combine to create textual coherence, which is separate from but dependent on cohesion. I worked with this framework for years on a corpus project. We annotated texts for every cohesive device we could find. What I learned quickly was that cohesion is not the same thing as flow. A text can be fully cohesive and still read like garbage. You can chain pronouns from end to end and the reader still won't understand what's going on. Cohesion is mechanical. Coherence is cognitive. The two interact, but they are not interchangeable.
Understanding Cohesion In English Mak Halliday
Halliday and Hasan's taxonomy breaks down into five major categories: reference, substitution, ellipsis, conjunction, and lexical cohesion. Each one operates differently in practice. Reference is the most common device. It includes personal reference (pronouns like he, she, it, they), demonstrative reference (this, that, these, those), and comparative reference (same, other, similar). When you write "The policy failed. It was poorly designed," the second sentence depends entirely on the first. The pronoun "it" picks up its meaning from "the policy." This is exophoric when the reference points outside the text to the situation, and endophoric when it points inside the text. Substitution replaces a word with another word rather than a phrase. Common examples use do, one, and so. "I don't like this model. Let me try another one." Here "one" substitutes for "model." This is distinct from ellipsis because substitution leaves a gap filled with a different word, while ellipsis simply removes the word entirely. You'll see this in technical writing more than anywhere else.
Ellipsis is the omission of something that can be recovered from context. "She can play piano, and he can [play] too." The bracketed verb is implied but not stated. This is frequent in conversation and often gets flagged as informal, but it's perfectly grammatical when the antecedent is clear. The risk is ambiguity. If your ellipsis lacks a clear anchor, the reader stalls. Conjunction connects clauses and sentences using logical markers. Additive (and, furthermore, similarly), adversative (but, however, on the other hand), causal (so, therefore, as a result), and temporal (then, after, meanwhile) are the main subtypes. Conjunctions don't refer to anything in particular. They signal the relationship between propositions. "The system crashed. As a result, we lost the data." The causal conjunction does the heavy lifting here. Lexical cohesion is often the most overlooked type. It operates through vocabulary choice: repetition, synonyms, collocation, and semantic fields. "The engineer inspected the bridge. The structural damage was severe." The link between "engineer" and "bridge" and "structural" creates cohesion without any explicit grammatical marker. This is where most non-native writers struggle because they default to repetition instead of variation.
Get the Full Details
Why People Mess This Up
The most common error I see is over-reliance on reference. Writers pile on pronouns until the text becomes impossible to track. "It happened because it was designed that way, and it wasn't tested, so when it failed, the team realized it should have been built differently." Four instances of "it" in one sentence. The reader has to backtrack and guess which "it" is which. By the end, the sentence is noise. Another frequent problem is inconsistency in substitution. You'll see texts that mix substitution and ellipsis unpredictably, or switch between them within the same paragraph. The reader's brain expects a pattern. When that pattern breaks, comprehension drops. It's a small thing but it adds up across longer documents. I ran into a specific edge case during a revision project for a technical manual. The original author used demonstrative references extensively. "This process ensures quality. This is critical." The word "this" appeared repeatedly with no clear antecedent. Was "this" referring to the process, the quality assurance, the entire workflow? I couldn't tell. The workaround was straightforward: replace each ambiguous demonstrative with a noun phrase. "This process" became "the calibration process," and "This is critical" became "The calibration step is critical." The revision took about forty minutes for a twelve-page document, but it eliminated every instance of that ambiguity. Readers later confirmed that the revised version was significantly easier to follow.
What Halliday's Framework Gets Wrong
For all its usefulness, the cohesion taxonomy has real limitations. It treats cohesive devices as discrete units when in practice they overlap constantly. A single sentence can contain reference, substitution, and conjunction at the same time. The framework doesn't account for this interaction well. More importantly, cohesion is culturally mediated. The devices that work in British academic writing may not transfer cleanly to American business communication or to non-native English contexts. Halliday and Hasan's examples are drawn from a fairly narrow set of registers. Modern technical and digital writing uses cohesive strategies that weren't in the original taxonomy. Acronyms, hashtags, hyperlinks, and code-switching all function as cohesive devices in ways the original model doesn't predict. There's also the issue of register bias. Formal prose rewards explicit cohesion. Casual conversation relies more on shared context and less on grammatical linking. If you apply the Halliday framework to a text message exchange, it will seem thin on cohesive devices even though the exchange is perfectly understandable. That's because conversational cohesion depends on prosody, timing, and social knowledge, not just grammar.
Practical Approaches for Checking Your Own Writing
The most efficient method I've found is reverse outlining. After you draft a section, read it sentence by sentence and label the cohesive device linking each sentence to the previous one. If you can't find a label, that's your problem spot. This takes about fifteen minutes for a standard essay and catches the majority of cohesion breaks. You don't need special software. A pencil and margin notes are sufficient. Another technique is to read your text aloud. Disfluent passages usually indicate missing or weak cohesive links. Your ear will catch what your eye skips over. I do this whenever I'm revising a document and something feels off but I can't pinpoint why. It's not foolproof but it's fast and surprisingly reliable. If you're working with a large volume of text, consider running a cohesion analysis script. There are open-source tools that can tag reference, substitution, and ellipsis automatically. They won't catch everything, but they give you a baseline. The output usually highlights pronoun density and conjunction frequency, which are the two metrics that correlate most strongly with readability issues.
When to Look Elsewhere
If you're analyzing discourse beyond the sentence level, Halliday's cohesion framework alone won't cut it. You'll need supplement it with something like conversation analysis for spoken interaction, or rhetorical structure theory for argumentative texts. Cohesion is necessary but not sufficient for good writing. Structure, argumentation, and audience awareness matter just as much. For legal and contract writing, the rules are stricter. Ambiguous reference is not acceptable. Every pronoun must have an unambiguous antecedent, and conjunctions need to specify the exact logical relationship. In these domains, cohesion isn't optional polish. It's a requirement. The original framework from Halliday and Hasan remains the foundation, but the field has moved past treating it as complete. New research in computational linguistics and discourse analysis has refined and extended the taxonomy considerably. The core insight holds: texts stick together because of deliberate linguistic choices. The details of how that works keep evolving.
